AI filmmaking
What a Short Film Made With AI Video Actually Costs
Every AI video tool advertises a monthly price. None of them advertise the number that decides your bill, which is how many attempts it takes you to get a shot you can use.
How these tools actually charge
Almost all of them meter generation rather than time. You buy credits, a clip costs some number of them, and the monthly fee is really a credit allowance with a login attached. Longer clips and higher resolutions cost more, usually more than proportionally.
The consequence is that your cost is not set by the price list. It is set by your hit rate — how many generations you burn per shot you keep.
The number nobody tells you
A beginner's hit rate is roughly one usable shot in eight. Someone who has learned to write a shot properly runs closer to one in three. That is not a small difference: on a film needing thirty shots, it is 240 generations against 90.
At any price per generation, the person who learned shot design pays a third of what the person who did not pays — for the same film. This is why craft is the cost-control measure, and why buying a bigger plan to fix a low hit rate is the expensive way round.
The four costs people do not budget for
- Regenerating for continuity, not quality. The shot is good; it just does not match the one before it. These are pure craft losses and they vanish once you fix continuity at the writing stage.
- Upscaling and export. Often metered separately from generation, and easy to miss until a final pass costs more than you expected.
- The month you do not use. Subscriptions renew whether or not you made anything. Three idle months is a real number on an annual view.
- Audio. Voice, music and effects are usually a different product with its own bill, and a silent film feels unfinished no matter how good the pictures are.
A realistic shape for a first short film
For a two to three minute piece, roughly thirty usable shots:
| Learning as you go | After some practice | |
|---|---|---|
| Usable shots needed | 30 | 30 |
| Generations per usable shot | ~8 | ~3 |
| Total generations | ~240 | ~90 |
| Months of subscription | 2–3 | 1 |
| Relative cost | High | Roughly a third |
Deliberately without currency figures: per-credit pricing moves constantly and any number printed here would be wrong within months. The ratio is the durable part, and the ratio is the point.
Generate at the lowest resolution that lets you judge the shot, and only upscale what survives into the edit. People routinely pay full price for high-resolution versions of clips they then cut.
Is it cheaper than filming it?
For a short with no crew, no location and no cast, yes — substantially, and that is a real change in what one person can attempt.
But the honest comparison is not against a film crew. It is against not making it at all, which is what most of these projects were before. The question worth asking is not whether AI video is cheaper than a camera, but whether what you get out is worth what you put in — and that depends almost entirely on whether you learned to direct it.
Where the money is actually saved
Not in the tool. In the three habits that raise your hit rate: writing the shot before generating it, keeping constants identical between shots in a scene, and abandoning a shot after three failed attempts instead of twelve.
That last one is the hardest and the most valuable. The shot you cannot get is telling you something about the scene, and rewriting it is nearly always cheaper than winning the argument.